Improving P300 and SCP-based Brain computer interfacing by spectral subtraction denoising

Abstract

A new denoising technique for preprocessing of P300 and Slow Cortical Potential (SCP)-based Brain computer interface data is proposed. This new technique adaptively removes the superimposed noise using a modified version of spectral subtraction method. A better performance is achieved especially when less number of electrodes is used which accordingly reduce weight and consumed power for portable BCI applications. Classification accuracy and bitrate estimate were used as quantitative performance measures. Results showed better performance when compared to preprocessing without denoising and with using the relevant and widely used wavelet shrinkage denoising method. Results proved the practical utility of this method and we suggest adding it to different BCI experiments.